Arm Neoverse: Powering the Next-Generation of High-Performance Computing

Article By : Stephen Las Marias

Arm's Neoverse platform and ecosystem can help foster innovation and growth with successful deployment in the hyperscale and enterprise cloud data centers.

India’s digital economy is in a stage of exciting growth. With over a billion mobile phones in use in the country and around 700 million internet subscribers, the opportunities for an ecosystem powered by digitalization are endless.

In fact, India now is one of the leaders in data consumption and generation worldwide. The outbreak of the COVID-19 pandemic in 2020 further accelerated the adoption of cloud computing in the country as enterprises sent employees to work from home and schools turned to online education. Add to this the demand for online services brought about by video streaming and gaming as people get to stay at home amid lockdowns and movement control orders, social media platforms, as well as increasing e-commerce activities.

All of these trends are fueling the growth of the country’s data center infrastructure industry. According to JLL India, India’s data center industry is expected to reach 1,007 MW by 2023, more than double its existing capacity of 447 MW.

“The growth of the digital economy is going to lead to the growth of India’s data center industry over the next few years,” said Eddie Ramirez, Sr. Director of Marketing, Infrastructure, at Arm. “The Ministry of Electronics and IT (MeitY) published a report last year saying that by 2025, there will be $4.9 billion spent on data centers within the country.”

Ramirez leads the go-to market and ecosystem team for Arm’s infrastructure line of business. “For us, infrastructure is everything in the data center, including the networks such as 5G that power data that goes across the world,” he said. “We are the group that’s looking at how to improve compute power for the infrastructure.”

In a recent webinar titled “Disrupting Cloud Data Centers with Arm Neoverse”, Ramirez discussed how Arm’s Neoverse platform and ecosystem can help foster innovation and growth with successful deployment in the hyperscale and enterprise cloud data centers. He also highlighted the comprehensive hardware and software ecosystem that enable and optimize customers’ application development and deployment on Arm-based infrastructure.

The Neoverse Platform

Conceptualizing the Neoverse platform, Ramirez said they “started with the simple question of ‘How do we build a platform that can get you more compute from the same power output?’.”

“If all these data centers are going to be built over the next five years in India, how do we scale the compute to use that space most efficiently? Every data center has a certain power footprint that they have to operate about,” he said.

That was the fundamental question that Arm addressed with the Neoverse platform. Designed specifically for infrastructure and cloud computing segments, Arm’s Neoverse platform—starting with the N1 and E1 released in 2019, followed by N2 and V1 released this year—is the foundation for the next generation cloud-to-edge infrastructure, delivering high-performance, secure, and scalable computing solutions along with a robust hardware and software ecosystem.

Since 2018 when Arm first announced Neoverse, the company has seen a wave of adoption throughout cloud-to-edge infrastructure. The rich diversity of hardware and software solutions that have come to market enabled by Neoverse-based compute are now deployed in cloud data centers, HPC systems, 5G networks, and out to edge gateways, providing cost savings, power efficiency, and compute performance gains.

“We are now seeing cloud service provides like AWS [Amazon Web Services] and Oracle adopting Arm and offering compute instances that are both high performing and offer costs advantages,” said Ramirez.

Designed by their Annapurna Labs team, AWS’s own server CPU called Graviton2 delivers 64 Arm Neoverse N1 cores on 7nm manufacturing technology. With Neoverse, AWS was able to demonstrate a 40% better price performance running on Graviton2-based compute instances than what they had before with the legacy architectures.

“That’s really significant because not only are they able to build their own processors, but they are also now more in control of their supply chain,” said Ramirez. “But to actually be able to pass on these very significant performance and cost savings to their end customer really puts them in a different class of cloud providers.”

AWS now has several EC2 compute instances running on Graviton2. Most recently, the company launched new extra-high memory X2gd instances which, in some cases, are providing over 80% better throughput compared to older X1 models.

“We were excited by the performance benefits that we at Arm are now shifting more of our EDA workflows to Graviton2.​ And we’re happy with the overall performance and TCO benefits we have achieved,” said Ramirez.

Another cloud service provider embracing Arm’s Neoverse platform is Oracle. “Oracle’s known for their database software, but they also have quite a significant presence in their Oracle Cloud. They launched their Arm-based cloud instances utilizing two socket servers equipped with Ampere Computing’s Altra 80-core CPUs for a total of 160 Arm Neoverse N1 cores per server. The systems include 1TB of memory and 2×50 Gbps networking. This powerful server allows customer flexibility to enable right size of compute and memory to support their needs,” explained Ramirez.

He said Oracle was the first to announce a penny per core-hour that customers can use—bringing the cost of compute down significantly for customers that are using the public cloud.

Enabling the Next-Generation of HPC

“One of the things that differentiates Neoverse from some of the x86-type architectures out there is that we focus our designs on single-threaded performance versus using this concept of multithreading, where different threads share the same core,” said Ramirez.

This brings a more predictable performance, according to him. “If you are using a public cloud on an Arm Neoverse core, you can be sure that your virtual machine is accessing the full core on its own—you are not time sharing with other customers,” Ramirez explained.

This also provides benefits from a security standpoint because you are isolated to that single core.

And then it is not just about the cores, but the interconnects. “Our Neoverse CPU cores combined with our Coherent Mesh Interconnect products enables superior performance for high core count systems,” said Ramirez.

Last but not the least, it is also about the generation uplift. “The other thing that we look at is how do we ensure that we deliver generation to generation performance improvements. With our newer roadmap on the N series and V series, we are now able to achieve 40 to 50% performance uplift. That’s really kind of been unheard of. And that level of performance improvement from one generation to the other is very unique to Arm,” said Ramirez.

The future CPU designs that will be powered by Arm Neoverse will enable continued scaling in data center performance.

“We are already seeing traction with our new Neoverse platforms. One example is MeitY in India has decided to license the Neoverse V1 platform for their exascale HPC CPU design. They join other HPC initiatives for Exascale computing project in Europe via SiPearl and in Asia through ETRI, who have also announced adoption of Neoverse V1,” said Ramirez.

Enabling an Ecosystem

“At Arm, we are working every day to ensure that software can easily be developed and deployed on Arm platforms,” said Ramirez. “We see a future where all of the world’s shared digital data will, at some point in its lifetime, be processed on Arm. To execute this vision requires significant investment in software and support for developers who write the code.

He noted that developers are also rapidly adopting cloud native software. “We have a significant footprint of OSS projects, independent software vendors already supporting Arm 64-bit architecture,” Ramirez said. “We were really excited to learn from Docker that there are now over 100,000 containers that are written for Arm processors that are on their site today.”

He added that the other part of cloud native is deploying CI/CD—continuous integration continuous development tools—to ensure that anything developers changed or modify, features they add to their software, get tested daily.

“One of the things that we have done to help spur that is the Works on Arm program, where we are offering CPU cycles—they could be virtualized or they could be bare metal servers—that developers can take advantage of for free, as part of their co-development process,” said Ramirez. “The ecosystem has come a long way on Arm and it also helps that we have partners, like AWS, who are contributing to the ecosystem, as well as several independent software vendors that have made their efforts to port and optimize on Arm.”

There are now several OEMs and ODMs offering Arm-based servers in India. “Companies like Foxconn, Wiwynn, and Gigabyte have deployed multiple skews of Ampere Altra-based servers,” said Ramirez. “We continue to see more OEMs engaging us every day. And we are also excited to work with local vendors in India who may be interested in supporting Arm-based servers as well.”

Innovations in the Pipeline

Arm is a company that focuses on relentless IP innovation. And one of the things that the company introduced earlier this year is the Armv9 architecture—its first major upgrade in a decade.

According to Ramirez, one of the improvements in v9 is security, enabling things like confidential compute. “This is where you can ensure that a customer’s user data is effectively protected not only within the processor, but even within the virtual machine or within the container application that runs on that processor,” he explains. “We are also introducing enhancements to performance. Part of v9 is our scalable vector technology. Vectors—sort of one-dimensional arrays of data—have been around since the first supercomputers. With Armv9 and the SVE2 upgrade, chip designers now have a lot more flexibility in the vector links that they want to deploy. This will all help with delivering higher performance for workloads like genomics, computer vision, VR, and even machine learning on CPU.”

India and Beyond

Guru Ganesan, President of Arm India, sees wide adoption of Arm technology in the Indian cloud computing and telecom space in the coming years.

“Public cloud end-user spending in India is forecasted to be over $4B in 2021. Large enterprises, medium businesses, and start-ups in India will see significant performance and costs benefits by moving to higher performance and power-efficient Arm-based CPUs in the cloud. Additionally, as companies become more conscious of the environmental impact, it is important to consider the energy efficiency of Arm-based computing. Our engagement with the Indian government on the HPC front is progressing well, with MeITY starting to develop an HPC processor based on Arm Neoverse Technology”

“We have done a few supercomputing projects, most notably is the Fugaku supercomputer in Japan, where we helped enable RIKEN to build the most powerful supercomputer in the world using Fujitsu’s A64FX CPU,” said Ramirez. “That has delivered almost 7.6 million cores of processing power, so we are very excited to see what we can do with entities like MeitY in India. Not only for the cloud space, but for the HPC space and academics, or companies using supercomputing power—we are hoping that we bring such high-performance solutions to the India market.”

Arm is also working with other countries beyond India, in projects including 5G network deployments.

“India’s telecom ecosystem consisting of the network operators as well as OEM’s, are actively pursuing development of modular and interoperable, best-in-class hardware and software elements, to build state of the art, scalable and manageable 5G networks,” said Ganesan. “Arm-based products, built on the concept of heterogenous compute, offer a complete set of solutions all the way from radio-unit to the core, to enable deployment of high-performance networks with the lowest TCO.”

“We’ve been working very closely with different countries in Southeast Asia on how to enable, for example, O-RAN initiatives. We’ve been a big part of O-RAN, and this will have a big impact in Southeast Asia, as they are now looking at deploying 5G networks in different countries,” said Ramirez.  “That has been a big initiative for us—to participate in those standard groups—to drive this open architecture for 5G networks.”

“With respect to Southeast Asia, opportunities abound both with AWS as Eddie mentioned and Oracle—as it is expanding its Arm instance presence in all regions globally,” said Amaresh Iyer, Senior Manager, Segment Marketing, Infrastructure BU at Arm. “And their pricing offer that they have on Arm instance in the OCI is a golden opportunity for a lot of developers, especially in countries like India and Southeast Asia, to take advantage of when it comes to testing, porting, and recategorizing their workloads on Arm, at a very low cost.”

Iyer noted that another big factor offered by Arm is the sustainable power-efficient processor. “In these countries where power is a big issue, having power-efficient datacenter infrastructure is very important. That’s very attractive to markets like India and countries in Southeast Asia,” he explained. “We see a lot of developments happening in 5G, Internet of Things—all of those market segments we also play in as part of the Arm IP infrastructure. And we have a global view—from edge to cloud—and Arm has an IP offering in each of those segments. These are extensive technology offerings that are secure, scalable, power efficient, and high-performance, and suitable for many different markets worldwide.”


Leave a comment